CN105528423A - Self-adaptive co-location pattern obtaining method and apparatus considering spatial instance distance weight - Google Patents

Abstract

The present invention provides a self-adaptive co-location pattern obtaining method and apparatus considering a spatial instance distance weight. The method comprises the steps of: establishing Thiessen polygons by using instances having the same spatial characteristics as target layers, and establishing a normalization instance dual table on basis of connection relationships of instances of the current target layer and other characteristic instances in a range of influence of the Thiessen polygons; performing distribution parameter calculation on a distance set of the normalization instance dual table, and obtaining normalization distance truncation parameters; establishing a distance feedback function by utilization of the normalization distance truncation parameters, and calculating a distance feedback value of a connection instance couple having any two different characteristics; and obtaining the final co-location pattern based on distance feedback function calculation on the basis of the distance feedback value. The defects of a conventional co-location pattern mining method in aspects of area adaptability, high efficiency and validity are solved.

Description

Take the self-adaptation coordination pattern acquiring method and apparatus of space instances distance weighting into account

Technical field

The present invention relates to spatial data mode excavation field, particularly relate to a kind of adaptive for spatial point factor data, and consider a coordination mode excavation method and apparatus for some key element adjacency weight.

Background technology

Spatial co-location patterns refers to the combination of a series of space characteristics, and the some example key element of these combinations occurs frequently in same place, and often characterize them spatially has certain general dependence.Coordination mode excavation is widely used in fields such as Species structure, mobile communication, public safety, environmental management, city plannings at present, the inherent mechanism that its result excavated can study space characteristics distribution further for scholar is given a clue, and also aid decision making person can carry out the anticipation of science.For city planning, someone finds that 1000 meters, a certain new city most school periphery does not have postal point, and utilize the facility data in Core Feature district, Beijing to carry out coordination mode excavation, its result shows that school and postal point belong to the same bit pattern that significance is 89%, based on this fact, what decision maker can be appropriate introduces school's periphery by postal service, to reduce movement of population radius, the wasting of resources of traffic pressure, the reduction each side of postponing.

Traditional coordination mode excavation method generally needs the distance threshold that user's sets itself is unified, improve even if wherein had some to expand to some extent with excavation target due to handled data type and done in the index of original distance metric and changed other index into, as distance threshold is changed into density threshold.But, this only to carry out the method for setpoint distance threshold value with user experience very unrealistic: region has specificity, Data distribution8 on region also has dividing of density, and distance threshold less than normal is difficult to dig out useful pattern, and distance threshold bigger than normal can cause pattern useless in a large number to produce.Although at present more existing improve one's methods and do not need user that distance threshold is set, but they generally utilize the method for arest neighbors thinking or iteration optimization to transform algorithm, be difficult to the understanding distance feature between example being had to the overall situation, and consider that the efficiency of overall annexation can the calculating of limit algorithm on large data sets.Secondly, classic method, based on the pop mode decision method of participation, does not consider the connection tightness degree between example in the calculation, namely distance weighting between example is not discussed on the impact of pattern popularity degree.And in fact, according to First Law of Geography, close things contact is tightr, so closely example relation, on the impact of pattern popularity degree size, is greater than remote example relation.Therefore, need to build the conspicuousness that a kind of novel effective counting system carrys out evaluation model.

In a word, document not relevant at present or disclosed method, when can consider distance weighting impact between area data density and example at the same time, efficiently find same bit pattern spatially fast.

Summary of the invention

The present invention is intended to solve above-described problem.An object of the present invention is to provide any one self-adaptation coordination pattern acquiring method and apparatus taking space instances distance weighting into account in a kind of overcoming the above problems.Be specially:

An aspect of of the present present invention provides a kind of self-adaptation coordination pattern acquiring method taking space instances distance weighting into account, comprising:

Using the example of the same space feature as target layer, build Thiessen polygon, based on example and the further feature example of current goal layer in the annexation of Thiessen polygon coverage, structure normalization example antithesis table;

Distribution parameter calculating is carried out to the distance set of normalization example antithesis table, obtains normalized cumulant Truncation Parameters;

Utilize normalized cumulant Truncation Parameters to construct distance feedback function, calculate the distance backoff values that the connection example of any two different characteristics is right;

Under described distance feedback threshold value, obtain the final same bit pattern calculated based on distance feedback function.

Method provided by the invention also has following features: the step of described structure normalization example antithesis table comprises:

For each space-like feature e _i, using the example as target layer, create Thiessen polygon figure;

Between the zone of influence of the object instance of delimiting based on Thiessen polygon figure, set up each example with further feature e _jthe incidence relation of example, and these examples associated are stored in two-dimentional Hash table table_instance to the distance with them _vorin; Wherein, table_instance _vor(e _i, e _j) be the orderly pattern (e of second order _i, e _j) Cell information, represent with e _ifor target layer, be characterized as e _jexample carry out being connected obtained row example information with target layer example, representation feature is e _ia kth example, Cell information comprises the orderly pattern (e of second order _i, e _j) connected example to the distance value dis of insPair and its correspondence;

By formula (1), calculate different characteristic connection example between normalized cumulant, replace the distance value of described two-dimentional Hash table:

$Nordis\left(p_k^{e_i},p_w^{e_j}\right)=\frac{dis\left(p_k^{e_i},p_w^{e_j}\right)-\min \left({\mathrm{Dis}}_{VP}\right)}{\max \left({\mathrm{Dis}}_{VP}\right)-\min \left({\mathrm{Dis}}_{VP}\right)}---\left(1\right),$

Wherein, dis (a, b) represents two example a, the Euclidean distance between b,

Dis _vPrepresent table_instance _vorin the right distance set of all example,

Min (T) and max (T) represents the minimum and maximal value of getting set T respectively;

Unordered process is carried out to described two-dimentional Hash table, obtains normalization example antithesis table step comprises:

The table_instance meeting j>i _vor(e _j, e _i) in the example of arbitrarily in order second mode Cell information order is exchanged after, obtain new two-dimentional Hash table

Meet j>i's by described two-dimentional Hash table newly cell information and former table_instance _vor(e _i, e _j) corresponding Cell information carries out asking and processing, and obtains final normalization example antithesis table

Method provided by the invention also has following features: instance number is greater than to the data set of 12000, utilize the normalized cumulant set in described normalization example antithesis table, under the prerequisite of level of signifiance α=0.05, carry out the parameter fitting of broad sense Extreme maximum distribution function, for χ ²the situation that in inspection, the goodness of fit is greater than 0.8, the normalized cumulant described in setting blocks as μ ± stderr; Wherein μ represents the location parameter of generalized extreme value maximum value fitting function, and stderr represents the residual error of matching; Or,

Other is not met to the situation of instance number and goodness of fit condition, using the average of normalized cumulant as cutoff value.

Method provided by the invention also has following features: utilize described normalized cumulant Truncation Parameters, builds distance feedback function, calculates the connection example pair of any two different characteristics distance backoff values:

$\mathrm{Re}ward\left(p_k^{e_i},p_w^{e_j}\right)=\left\{\begin{array}{cc}{\left(\frac{l_{close}-Nordis\left(p_k^{e_i},p_w^{e_j}\right)}{l_{close}}\right)}^2& Nordis\left(p_k^{e_i},p_w^{e_j}\right)<l_{close}\\ -{\left(\frac{Nordis\left(p_k^{e_i},p_w^{e_j}\right)-l_{far}}{1-l_{far}}\right)}^2& Nordis\left(p_k^{e_i},p_w^{e_j}\right)>l_{far}\\ 0& l_{close}\&le;Nordis\left(p_k^{e_i},p_w^{e_j}\right)\&le;l_{far}\end{array}\right.---\left(2\right),$

Wherein, l _close=μ-stderr, and l _far=μ+stderr.

Method provided by the invention also has following features: under described distance feedback threshold value, and the step obtaining the final same bit pattern based on the calculating of distance feedback function comprises:

With the feature corresponding to the column locations of described normalization example antithesis table to { e _i, e _jbe initial candidate peers pattern c, the following sub-step 1-5 of iteration, until | c| rank are not greater than the member of distance feedback threshold value with bit pattern, wherein, and j>i,

When | during c|=2, omit sub-step 1-2, directly jump to sub-step 3;

Sub-step 1: for candidate pattern c={e _i... e _k, according to the example pair in described normalization example antithesis table, obtain the group example insClique of candidate pattern _c, wherein, the feature comprised in candidate pattern c arranges according to lexcographical order, namely meets k>=j>i, insClique _crefer to based on Thiessen polygon, spatially have the combination of the example of the different characteristic of full annexation, namely roll into a ball any two examples pair extracted in example, can respective items be found in normalization example antithesis table;

Sub-step 2: according to formula (3), asks any two the second mode { e calculating and comprise in candidate pattern c _i, e _jexample pair at insClique _cdistance backoff values under constraint:

${\mathrm{Reward}}_c\left(p_k^{e_i},p_w^{e_j}\right)=\left\{\begin{array}{cc}\mathrm{Re}ward\left(p_k^{e_i},p_w^{e_j}\right)& \left(p_k^{e_i},p_w^{e_j}\right)\&Element;{\mathrm{\&pi;}}_{\left\{e_i,e_j\right\}}\left({\mathrm{insClique}}_c\right)\\ -\left|\right|\mathrm{Re}ward\left(p_k^{e_i},p_w^{e_j}\right)\left|\right|& \left(p_k^{e_i},p_w^{e_j}\right)\&NotElement;{\mathrm{\&pi;}}_{\left\{e_i,e_j\right\}}\left({\mathrm{insClique}}_c\right)\end{array}\right.---\left(3\right),$

Wherein, represent and ask for second mode { e _i, e _jat insClique _con projection;

Sub-step 3: by formula (4), asks and calculates second mode { e _i, e _jat c={e _i... e _kconstraint under distance backoff values:

$\begin{array}{c}{\mathrm{Reward}}_c\left(e_i,e_j\right)=\\ avg\left\{{\mathrm{Reward}}_c\left(p_k^{e_i},p_w^{e_j}\right)|\left(p_k^{e_i},p_w^{e_j}\right)\&Element;table\_{\mathrm{instance}}_{\stackrel{\&OverBar;}{vor}}\left(e_i,e_j\right).insPair\right\}\end{array}---\left(4\right),$

Wherein, represent second mode { e _i, e _jall examples in corresponding normalization example antithesis table are to set, avg represents and asks calculation mean value;

Sub-step 4: by formula (5), ask calculation pattern c={e _i... e _kthe popularity degree calculated based on distance feedback function:

Reward _c＝min _i＜j≤k{Reward _c(e _i,e _j)}(5)，

Being judged to be final by being greater than to the candidate pattern of set a distance feedback threshold value θ | the same bit pattern in c| rank, namely final same bit pattern need meet Reward _c> θ;

Sub-step 5: with | c| rank with bit pattern for foundation, by front | the same bit pattern that c|-1 position has same characteristic features is asked also, obtains | c|+1 rank candidate peers pattern c ', makes c=c '.

Another aspect of the present invention additionally provides a kind of self-adaptation coordination pattern acquiring device taking space instances distance weighting into account, comprising:

Normalization example antithesis table builds module, for using the example of the same space feature as target layer, build Thiessen polygon, based on example and the further feature example of current goal layer in the annexation of Thiessen polygon coverage, structure normalization example antithesis table;

Normalized cumulant Truncation Parameters acquisition module, for carrying out distribution parameter calculating to the distance set of normalization example antithesis table, obtains normalized cumulant Truncation Parameters;

Example is adjusted the distance backoff values computing module, for utilizing normalized cumulant Truncation Parameters to construct distance feedback function, calculates the distance backoff values that the connection example of any two different characteristics is right;

Coordination pattern acquiring module, under described distance feedback threshold value, obtains the final same bit pattern calculated based on distance feedback function.

Device provided by the invention also has following features: described normalization example antithesis table build module specifically for:

Thiessen polygon creating unit, for for each space-like feature e _i, using the example as target layer, create Thiessen polygon figure;

Two dimension Hash table creating unit, between the zone of influence of object instance delimited based on Thiessen polygon figure, sets up each example with further feature e _jthe incidence relation of example, and these examples associated are stored in two-dimentional Hash table table_instance to the distance with them _vorin; Wherein, table_instance _vor(e _i, e _j) be the orderly pattern (e of second order _i, e _j) Cell information, represent with e _ifor target layer, be characterized as e _jexample carry out being connected obtained row example information with target layer example, representation feature is e _ia kth example, Cell information comprises the orderly pattern (e of second order _i, e _j) connected example to the distance value dis of insPair and its correspondence;

Normalized cumulant computing unit, for the connection example that calculates different characteristic between normalized cumulant, replace the distance value of described two-dimentional Hash table, see formula (1):

$Nordis\left(p_k^{e_i},p_w^{e_j}\right)=\frac{dis\left(p_k^{e_i},p_w^{e_j}\right)-\min \left({\mathrm{Dis}}_{VP}\right)}{\max \left({\mathrm{Dis}}_{VP}\right)-\min \left({\mathrm{Dis}}_{VP}\right)}---\left(1\right),$

Wherein, dis (a, b) represents two example a, the Euclidean distance between b,

Dis _vPrepresent table_instance _vorin the right distance set of all example,

Min (T) and max (T) represents the minimum and maximal value of getting set T respectively;

Unordered processing unit, for carrying out unordered process to described two-dimentional Hash table, obtains normalization example antithesis table

Described unordered processing unit, specifically for:

The table_instance meeting j>i _vor(e _j, e _i) in the example of arbitrarily in order second mode Cell information order is exchanged after, obtain new two-dimentional Hash table

Meet j>i's by described two-dimentional Hash table newly cell information and former table_instance _vor(e _i, e _j) corresponding Cell information carries out asking and processing, and obtains final normalization example antithesis table

Device provided by the invention also has following features: described normalized cumulant Truncation Parameters acquisition module specifically for:

Instance number is greater than to the data set of 12000, utilizes the normalized cumulant set in described normalization example antithesis table, under the prerequisite of level of signifiance α=0.05, carry out the parameter fitting of broad sense Extreme maximum distribution function, for χ ²the situation that in inspection, the goodness of fit is greater than 0.8, the normalized cumulant described in setting blocks as μ ± stderr; Wherein μ represents the location parameter of generalized extreme value maximum value fitting function, and stderr represents the residual error of matching; Or,

Other is not met to the situation of instance number and goodness of fit condition, using the average of normalized cumulant as cutoff value.

Device provided by the invention also has following features: described example adjust the distance backoff values computing module specifically for, calculate the connection example pair of any two different characteristics distance backoff values, see formula (2):

$\mathrm{Re}ward\left(p_k^{e_i},p_w^{e_j}\right)=\left\{\begin{array}{cc}{\left(\frac{l_{close}-Nordis\left(p_k^{e_i},p_w^{e_j}\right)}{l_{close}}\right)}^2& Nordis\left(p_k^{e_i},p_w^{e_j}\right)<l_{close}\\ -{\left(\frac{Nordis\left(p_k^{e_i},p_w^{e_j}\right)-l_{far}}{1-l_{far}}\right)}^2& Nordis\left(p_k^{e_i},p_w^{e_j}\right)>l_{far}\\ 0& l_{close}\&le;Nordis\left(p_k^{e_i},p_w^{e_j}\right)\&le;l_{far}\end{array}\right.---\left(2\right),$

Wherein, l _close=μ-stderr, and l _far=μ+stderr.

Device provided by the invention also has following features: described coordination pattern acquiring module, specifically for:

With the feature corresponding to the column locations of described normalization example antithesis table to { e _i, e _jbe initial candidate peers pattern c, the following sub-step 1-5 of iteration, until | c| rank are not greater than the member of distance feedback threshold value with bit pattern, wherein, and j>i,

When | during c|=2, omit sub-step 1-2, directly jump to sub-step 3;

Sub-step 1: for candidate pattern c={e _i... e _k, according to the example pair in described normalization example antithesis table, obtain the group example insClique of candidate pattern _c, wherein, the feature comprised in candidate pattern c arranges according to lexcographical order, i.e. k>=j>i, insClique _crefer to based on Thiessen polygon, spatially have the combination of the example of the different characteristic of full annexation, namely roll into a ball any two examples pair extracted in example, can respective items be found in normalization example antithesis table;

Sub-step 2: according to formula (3), asks any two the second mode { e calculating and comprise in candidate pattern c _i, e _jexample pair at insClique _cdistance backoff values under constraint:

${\mathrm{Reward}}_c\left(p_k^{e_i},p_w^{e_j}\right)=\left\{\begin{array}{cc}\mathrm{Re}ward\left(p_k^{e_i},p_w^{e_j}\right)& \left(p_k^{e_i},p_w^{e_j}\right)\&Element;{\mathrm{\&pi;}}_{\left\{e_i,e_j\right\}}\left({\mathrm{insClique}}_c\right)\\ -\left|\right|\mathrm{Re}ward\left(p_k^{e_i},p_w^{e_j}\right)\left|\right|& \left(p_k^{e_i},p_w^{e_j}\right)\&NotElement;{\mathrm{\&pi;}}_{\left\{e_i,e_j\right\}}\left({\mathrm{insClique}}_c\right)\end{array}\right.---\left(3\right),$

Wherein, represent and ask for second mode { e _i, e _jat insClique _con projection;

Sub-step 3: by formula (4), asks and calculates second mode { e _i, e _jat c={e _i... e _kconstraint under distance backoff values:

$\begin{array}{c}{\mathrm{Reward}}_c\left(e_i,e_j\right)=\\ avg\left\{{\mathrm{Reward}}_c\left(p_k^{e_i},p_w^{e_j}\right)|\left(p_k^{e_i},p_w^{e_j}\right)\&Element;table\_{\mathrm{instance}}_{\stackrel{\&OverBar;}{vor}}\left(e_i,e_j\right).insPair\right\}\end{array}---\left(4\right),$

Wherein, represent second mode { e _i, e _jall examples in corresponding normalization example antithesis table are to set, avg represents and asks calculation mean value;

Sub-step 4: by formula (5), ask calculation pattern c={e _i... e _kthe popularity degree calculated based on distance feedback function:

Reward _c＝min _i＜j≤k{Reward _c(e _i,e _j)}(5)，

Being judged to be final by being greater than to the candidate pattern of set a distance feedback threshold value θ | the same bit pattern in c| rank, namely final same bit pattern need meet Reward _c> θ;

Sub-step 5: with | c| rank with bit pattern for foundation, by front | the same bit pattern that c|-1 position has same characteristic features is asked also, obtains | c|+1 rank candidate peers pattern c ', makes c=c '.

With the subset that bit pattern is one group of space characteristics, represent that their example set is combined in and spatially occur frequently.In existing correlation technique, when being difficult to consider distance weighting impact between area data density and example at the same time, efficiently find same bit pattern spatially fast.

The present invention is directed to this problem, propose a kind of based on Thiessen polygon and the coordination pattern mining mode apart from feedback function, utilize the characteristic of Thiessen polygon, the Domain relation between different characteristic example is established in the range of influence of example, and on this basis, obtain the very big distance statistics feature between example, then characteristic parameter is used for the distance feedback function building candidate pattern, carrys out the popularity degree of measurement pattern.

The advantage of the method is: the operation 1) eliminating user's preset distance threshold value, has evaded the uncertainty carrying out excavating on zone of ignorance;

2) distance weighting is considered between example on the impact of pattern popularity degree, more realistic application scenarios;

3) use of Thiessen polygon, greatly reduces example to the calculated amount connected, operation efficiency is significantly improved.

Read the following description for exemplary embodiment with reference to accompanying drawing, other property feature of the present invention and advantage will become clear.

Accompanying drawing explanation

To be incorporated in instructions and the accompanying drawing forming a part for instructions shows embodiments of the invention, and together with the description for explaining principle of the present invention.In the drawings, similar Reference numeral is used for key element like representation class.Accompanying drawing in the following describes is some embodiments of the present invention, instead of whole embodiment.For those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can be obtained according to these accompanying drawings.

A kind of process flow diagram taking the self-adaptation coordination pattern acquiring method of space instances distance weighting into account that Fig. 1 provides for embodiments of the invention one;

Fig. 2 is the method schematic diagram building normalization example antithesis table;

Fig. 3 is from second order candidate peers pattern, obtains the iterative process figure of the final same bit pattern based on the calculating of distance feedback function;

A kind of structural representation taking the self-adaptation coordination pattern acquiring device of space instances distance weighting into account that Fig. 4 provides for embodiments of the invention two.

Embodiment

For making the object of the embodiment of the present invention, technical scheme and advantage clearly, below in conjunction with the accompanying drawing in the embodiment of the present invention, technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is the present invention's part embodiment, instead of whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art, not making other embodiments all obtained under creative work prerequisite, belong to the scope of protection of the invention.It should be noted that, when not conflicting, the embodiment in the application and the feature in embodiment can combination in any mutually.

In order to better, the technical scheme that embodiments of the invention provide is set forth, first following concept is described:

Thiessen polygon: the continuous polygon geometry that the perpendicular bisector connecting the straight line of two abutment points by a group forms.It divides plane according to most proximity principle, and each point is associated with its arest neighbors region.

The abbreviation of SHP data type: shapefile is the open form of a kind of spatial data of ESRI company exploitation.

Orderly pattern: refer to a series of vector having the space characteristics of sequence.

Generalized extreme value distribution: the curvilinear characteristic meeting the fat tail in peak, be an important branch of extreme value theory, unified Gumbel, Fr é chet and Weibull tri-kinds of Model of extreme distributions, for the application scenarios that analog variable is extreme value sequence, meteorology, economics are widely used.This distribution function relates to three key parameters, i.e. location parameter μ, scale parameter σ, and form parameter k.

Below in conjunction with accompanying drawing, embodiments of the invention one are described.

The invention provides a kind of self-adaptation coordination pattern acquiring method taking space instances distance weighting into account, be input as the point-like SHP data P={p of theorem in Euclid space ₁, p ₂... p _n, and the space characteristics type E={e involved by these data ₁, e ₂... e _m, wherein, each some example comprises numbering, type and X, Y-coordinate (in order to emphasize the space characteristics of example, below will be characterized as e _ia jth case representation be ).The flow process of the method as shown in Figure 1, comprising:

Step 101, using the example of the same space feature as target layer, build Thiessen polygon, based on example and the further feature example of current goal layer in the annexation of Thiessen polygon coverage, structure normalization example antithesis table;

The Construction Idea of normalization example antithesis table comes from a natural character: most of object spatially can be in the equilibrium point of the field force that they are formed.Two types (A, B) if object spatially there is certain relation, that object belonging to type A is more prone to the impact of the object being subject to the maximum category-B type of its periphery field force, and vice versa.Can infer thus, if A, category-B type attracts each other, and so their example can be distributed in the core area of range of influence each other mostly, otherwise then can be distributed in the edge zone that field force is the weakest each other.Because its attended operation is limited in the coverage of single instance, therefore its linking number compares traditional method, has and reduces significantly.This step relates to three sub-steps:

Sub-step 1: utilize Thiessen polygon to determine the domain object of all feature corresponding instances.For each space-like feature e _i, using the example under this feature as target layer, create Thiessen polygon figure.Like this, each example the example of relevant further feature can be expressed as:

$VP\left(p_j^{e_i}\right)=\left\{p\&Element;C_P^{P^{e_i}}|dis\left(p,p_k^{e_i}\right)\&le;dis\left(p,p_w^{e_i}\right)forallk\&NotEqual;w\right\}---\left(1\right)$

In above formula, can be calculated by overlay analysis in Geographic Information System; representation feature is non-e _iexample collection; Dis (a, b) represents the Euclidean distance of two space instances a, b.

Sub-step 2: by the example obtained in previous step to and distance value between them be stored in the two-dimentional Hash table of m*m, use table_instance _vorrepresent.Table_instance _vor(e _i, e _j) namely represent with e _ifor target layer, be characterized as e _jexample carry out being connected obtained row example information with target layer example, hereinafter referred to as the orderly pattern (e of second order _i, e _j) Cell information.Cell information comprises the orderly pattern (e of second order _i, e _j) connected example to the distance value dis of insPair and its correspondence.

By formula (2), calculate different characteristic connection example between normalized cumulant, replace the distance value of described two-dimentional Hash table.

$Nordis\left(p_k^{e_i},p_w^{e_j}\right)=\frac{dis\left(p_k^{e_i},p_w^{e_j}\right)-\min \left({\mathrm{Dis}}_{VP}\right)}{\max \left({\mathrm{Dis}}_{VP}\right)-\min \left({\mathrm{Dis}}_{VP}\right)}---\left(2\right)$

In above formula, dis (a, b) represents two example a, the Euclidean distance between b; Dis _vPrepresent table_instance _vorin the right distance set of all example; Min (T) and max (T) represents the minimum and maximal value of getting set T respectively.

Sub-step 3: by table_instance _vorin the example of orderly pattern unordered to becoming, namely the table_instance meeting j>i _vor(e _j, e _i) in the example of arbitrarily in order second mode Cell information order is exchanged after, obtain new two-dimentional Hash table then, meet j>i's by new two-dimentional Hash table cell information and former table_instance _vor(e _i, e _j) corresponding Cell information carries out asking and processing, and obtains final normalization example antithesis table

In order to the process of clearer explanation step 101, Fig. 2 gives patterned example.Series of features is had to be E={A, B in hypothesis space X ..., e _texample P={A ₁, A ₂... A ₅, B ₁... B ₉, C ₁... C ₅... }, the Thiessen polygon (for the purpose of simplifying the description, being only extract example with shadow region) that it is target figure layer building that Fig. 2 (a) illustrates with the example of feature A, wherein, example A ₅correspondence connect example be B ₄, B ₈, B ₇and C ₃.Example is finished to after connection and range normalization process to all features, defines the two-dimentional Hash table shown in Fig. 2 (b).Finally, to table_instance _vorcarry out disordering process, form Fig. 3's (c)

Step 102, distribution parameter calculating is carried out to the distance set of normalization example antithesis table, obtain normalized cumulant Truncation Parameters;

This step is by the Statistical Distribution of the distance of the further feature example in computer memory example and its coverage, obtain two " Truncation Parameters " referring to the general packing density feature in this region, in order to replace the distance threshold assignment procedure of traditional coordination mode excavation.

In the building process of normalization example antithesis table, asking for the example of instant example near target signature based on Thiessen polygon, is the evaluation strategy of a kind of region maximum value, therefore, can infer " normalized cumulant set (is denoted as Nordis _vP) " distribution be more prone to generalized extreme value distribution rule.In fact, the present invention utilizes Pekinese's urban facilities point-like data to carry out the experiment of subregion coordination mode excavation, the Nordis in 8 the continuous level regions of data volume all more than 1.2 ten thousand _vP, at the χ of level of signifiance α=0.05 ²during matching detects, all meet the Statistical Distribution of broad sense maximum value, the goodness of fit, all more than 0.89, is up to 0.9839, meets the statistical fit requirement of mankind's activity phenomenon.Thus, exist in, those have the feature pair of potential proximity relations, usually comprise the more heterogeneous normalized cumulant value less than normal to " the matching peak value μ ± stderr (α=0.05) of variable ", otherwise, then comprise more heterogeneous for matching peak value normalized cumulant value bigger than normal.Wherein, μ represents the location parameter of generalized extreme value maximum value fitting function, and stderr represents the residual error of matching.

According to above-mentioned reasoning and described experimental result, be greater than 12000 in instance number, and the χ of level of significance α=0.05 ²during matching detects, if the goodness of fit is greater than 0.8, the present invention using two boundary values of μ ± stderr (α=0.05) as Truncation Parameters, these two values are [0,1] distance axis subdivision is three parts: 1) > μ+stderr (α=0.05), refers to remote relation; 2) < μ-stderr (α=0.05), refers to closely relation; 3) [μ-stderr, μ+stderr], characterizes the maximum probability that occurs in this interval of normalized cumulant, reflection be the average characteristics distributed.Other is not met to the situation of instance number and goodness of fit condition, using the average of normalized cumulant as cutoff value.

Step 103, utilize normalized cumulant Truncation Parameters to construct distance feedback function, calculate the distance backoff values that the connection example of any two different characteristics is right;

According to First Law of Geography, the closely example pair of relation, to the contribution of pattern popularity degree, is greater than the example pair of remote relation.Based on this setting, the present invention constructs a monotonic decreasing function, sees formula (3), in order to calculate two dissimilar examples pair distance backoff values, the span of this function is [-1,1].

$\mathrm{Re}ward\left(p_k^{e_i},p_w^{e_j}\right)=\left\{\begin{array}{cc}{\left(\frac{l_{close}-Nordis\left(p_k^{e_i},p_w^{e_j}\right)}{l_{close}}\right)}^2& Nordis\left(p_k^{e_i},p_w^{e_j}\right)<l_{close}\\ -{\left(\frac{Nordis\left(p_k^{e_i},p_w^{e_j}\right)-l_{far}}{1-l_{far}}\right)}^2& Nordis\left(p_k^{e_i},p_w^{e_j}\right)>l_{far}\\ 0& l_{close}\&le;Nordis\left(p_k^{e_i},p_w^{e_j}\right)\&le;l_{far}\end{array}\right.---\left(3\right)$

In above formula, l _close=μ-stderr, and l _far=μ+stderr.

Step 104, under given distance feedback threshold value, obtain the final same bit pattern calculated based on distance feedback function.

This step with the feature corresponding to the column locations of normalization example antithesis table to { e _i, e _jbe initial candidate peers pattern c, iteration sub-step 1-5, until | c| rank are not greater than the member of distance feedback threshold value with bit pattern, and the iterative process of this step is as shown in Figure 3.Wherein, j>i.When | during c|=2, omit sub-step 1-2, directly jump to sub-step 3.

Sub-step 1: for candidate pattern c={e _i... e _k, according to the example pair in described normalization example antithesis table, obtain the group example insClique of candidate pattern _c.Wherein, the feature comprised in candidate pattern c arranges according to lexcographical order, i.e. k>=j>i, insClique _crefer to based on Thiessen polygon, spatially have the combination of the example of the different characteristic of full annexation, namely roll into a ball any two examples pair extracted in example, can respective items be found in normalization example antithesis table.

Sub-step 2: according to formula (3), asks any two the second mode { e calculating and comprise in candidate pattern c _i, e _jexample pair at insClique _cdistance backoff values under constraint.The example pair that the group's of participation example builds, retains the distance backoff values that it is original; The example pair forming group's example can not be participated in, a reverse effect is had to the popularity degree of pattern, and this backward-acting, defined by the absolute value negative asking for its distance backoff values.

${\mathrm{Reward}}_c\left(p_k^{e_i},p_w^{e_j}\right)=\left\{\begin{array}{cc}\mathrm{Re}ward\left(p_k^{e_i},p_w^{e_j}\right)& \left(p_k^{e_i},p_w^{e_j}\right)\&Element;{\mathrm{\&pi;}}_{\left\{e_i,e_j\right\}}\left({\mathrm{insClique}}_c\right)\\ -\left|\right|\mathrm{Re}ward\left(p_k^{e_i},p_w^{e_j}\right)\left|\right|& \left(p_k^{e_i},p_w^{e_j}\right)\&NotElement;{\mathrm{\&pi;}}_{\left\{e_i,e_j\right\}}\left({\mathrm{insClique}}_c\right)\end{array}\right.---\left(3\right)$

In above formula, represent and ask for second mode { e _i, e _jat insClique _con projection.

Sub-step 3: by formula (4), asks and calculates second mode { e _i, e _jat c={e _i... e _kconstraint under distance backoff values.Remote and closely example pair is scrupled in the calculating of this value simultaneously, on the impact of pattern popularity degree, therefore on the significance of pattern judges, than the epidemic index computing method of traditional only dependence " minimum value of the ratio of the example projection number in all single feature of pattern belonging to group's example and the example number of its same characteristic features ", more convincing.

$\begin{array}{c}{\mathrm{Reward}}_c\left(e_i,e_j\right)=\\ avg\left\{{\mathrm{Reward}}_c\left(p_k^{e_i},p_w^{e_j}\right)|\left(p_k^{e_i},p_w^{e_j}\right)\&Element;table\_{\mathrm{instance}}_{\stackrel{\&OverBar;}{vor}}\left(e_i,e_j\right).insPair\right\}\end{array}---\left(4\right)$

In above formula, represent second mode { e _i, e _jall examples in corresponding normalization example antithesis table are to set, avg represents and asks calculation mean value.

Sub-step 4: by formula (5), ask calculation pattern c={e _i... e _kthe popularity degree calculated based on distance feedback function.

Reward _c＝min _i＜j≤k{Reward _c(e _i,e _j)}(5)

As can be seen from formula (5), the span of this value is (-1,1), therefore user can simulate the setting means of popularity threshold, for the distance backoff values setting lower limit of pattern, to obtain the coordination mode excavation result of level, be about to be greater than and be judged to be final to the candidate pattern of set a distance feedback threshold value θ | the same bit pattern in c| rank.

Sub-step 5: with | c| rank with bit pattern for foundation, by front | the same bit pattern that c|-1 position has same characteristic features is asked also, obtains | c|+1 rank candidate peers pattern c ', makes c=c '.

Fig. 4 is the self-adaptation coordination pattern acquiring device taking space instances distance weighting into account provided by the invention.Fig. 4 shown device, comprising:

Normalization example table builds module 401, for being target layer by the example of the same space feature, build Thiessen polygon, based on example and the further feature example of current goal layer in the annexation of Thiessen polygon coverage, structure normalization example antithesis table;

Normalized cumulant Truncation Parameters acquisition module 402, for carrying out distribution parameter calculating to the distance set of normalization example antithesis table, obtains normalized cumulant Truncation Parameters;

Example is adjusted the distance backoff values computing module 403, and the distance feedback function constructed for utilizing normalized cumulant Truncation Parameters, calculates the distance backoff values that the connection example of any two different characteristics is right;

Coordination pattern acquiring module 404, under described distance feedback threshold value, obtains the final same bit pattern calculated based on distance feedback function.

Wherein, described structure module 401 specifically for:

Thiessen polygon creating unit, for for each space-like feature e _i, using the example as target layer, create Thiessen polygon figure;

Two dimension Hash table creating unit, between the zone of influence of object instance delimited based on Thiessen polygon figure, sets up each example with further feature e _jthe incidence relation of example, and these examples associated are stored in two-dimentional Hash table table_instance to the distance with them _vorin; Wherein, table_instance _vor(e _i, e _j) be the orderly pattern (e of second order _i, e _j) Cell information, represent with e _ifor target layer, be characterized as e _jexample carry out being connected obtained row example information with target layer example, Cell information comprises the orderly pattern (e of second order _i, e _j) connected example to the distance value dis of insPair and its correspondence;

Normalized cumulant computing unit, for the connection example that calculates different characteristic between normalized cumulant, replace the distance value of described two-dimentional Hash table, see formula (1):

$Nordis\left(p_k^{e_i},p_w^{e_j}\right)=\frac{dis\left(p_k^{e_i},p_w^{e_j}\right)-\min \left({\mathrm{Dis}}_{VP}\right)}{\max \left({\mathrm{Dis}}_{VP}\right)-\min \left({\mathrm{Dis}}_{VP}\right)}---\left(1\right),$

In above formula, dis (a, b) represents two example a, the Euclidean distance between b,

Dis _vPrepresent table_instance _vorin the right distance set of all example,

Min (T) and max (T) represents the minimum and maximal value of getting set T respectively;

Unordered processing unit, for carrying out unordered process to described two-dimentional Hash table, obtains normalization example antithesis table

Wherein, described unordered processing unit, specifically for:

The table_instance meeting j>i _vor(e _j, e _i) in the example of arbitrarily in order second mode Cell information order is exchanged after, obtain new two-dimentional Hash table

Meet j>i's by described two-dimentional Hash table newly in Cell information and former table_instance _vor(e _i, e _j) corresponding Cell information carries out asking and processing, and obtains final normalization example antithesis table

Wherein, described normalization Truncation Parameters acquisition module 402, specifically for:

Instance number is greater than to the data set of 12000, utilizes the normalized cumulant set in described normalization example antithesis table, under the prerequisite of level of signifiance α=0.05, carry out the parameter fitting of broad sense Extreme maximum distribution function, for χ ²the situation that in inspection, the goodness of fit is greater than 0.8, the normalized cumulant described in setting blocks as μ ± stderr; Wherein μ represents the location parameter of generalized extreme value maximum value fitting function, and stderr represents the residual error of matching; Or,

Other is not met to the situation of instance number and goodness of fit condition, using the average of normalized cumulant as cutoff value.

Wherein, described example is adjusted the distance backoff values computing module 403, specifically for, calculate the connection example pair of any two different characteristics distance backoff values, see formula (2):

$\mathrm{Re}ward\left(p_k^{e_i},p_w^{e_j}\right)=\left\{\begin{array}{cc}{\left(\frac{l_{close}-Nordis\left(p_k^{e_i},p_w^{e_j}\right)}{l_{close}}\right)}^2& Nordis\left(p_k^{e_i},p_w^{e_j}\right)<l_{close}\\ -{\left(\frac{Nordis\left(p_k^{e_i},p_w^{e_j}\right)-l_{far}}{1-l_{far}}\right)}^2& Nordis\left(p_k^{e_i},p_w^{e_j}\right)>l_{far}\\ 0& l_{close}\&le;Nordis\left(p_k^{e_i},p_w^{e_j}\right)\&le;l_{far}\end{array}\right.---\left(2\right),$

In above formula, l _close=μ-stderr, and l _far=μ+stderr.

Wherein, described coordination pattern acquiring module 404, specifically for:

With the feature corresponding to the column locations of described normalization example antithesis table to { e _i, e _jbe initial candidate peers pattern c, the following sub-step 1-5 of iteration, until | c| rank are not greater than the member of distance feedback threshold value with bit pattern, wherein, and j>i,

When | during c|=2, omit sub-step 1-2, directly jump to sub-step 3;

Sub-step 1: for candidate pattern c={e _i... e _k, according to the example pair in described normalization example antithesis table, obtain the group example insClique of candidate pattern _c, wherein, the feature comprised in candidate pattern c arranges according to lexcographical order, i.e. k>=j>i, insClique _crefer to based on Thiessen polygon, spatially have the combination of the example of the different characteristic of full annexation, namely roll into a ball any two examples pair extracted in example, can respective items be found in normalization example antithesis table;

Sub-step 2: according to formula (3), asks any two the second mode { e calculating and comprise in candidate pattern c _i, e _jexample pair at insClique _cdistance backoff values under constraint:

${\mathrm{Reward}}_c\left(p_k^{e_i},p_w^{e_j}\right)=\left\{\begin{array}{cc}\mathrm{Re}ward\left(p_k^{e_i},p_w^{e_j}\right)& \left(p_k^{e_i},p_w^{e_j}\right)\&Element;{\mathrm{\&pi;}}_{\left\{e_i,e_j\right\}}\left({\mathrm{insClique}}_c\right)\\ -\left|\right|\mathrm{Re}ward\left(p_k^{e_j}\right)\left|\right|& \left(p_k^{e_i},p_w^{e_j}\right)\&NotElement;{\mathrm{\&pi;}}_{\left\{e_i,e_j\right\}}\left({\mathrm{insClique}}_c\right)\end{array}\right.---\left(3\right),$

In above formula, represent and ask for second mode { e _i, e _jat insClique _con projection;

Sub-step 3: by formula (4), asks and calculates second mode { e _i, e _jat c={e _i... e _kconstraint under distance backoff values:

$\begin{array}{c}{\mathrm{Reward}}_c\left(e_i,e_j\right)=\\ avg\left\{{\mathrm{Reward}}_c\left(p_k^{e_i},p_w^{e_j}\right)|\left(p_k^{e_i},p_w^{e_j}\right)\&Element;table\_{\mathrm{instance}}_{\stackrel{\&OverBar;}{vor}}\left(e_i,e_j\right).insPair\right\}\end{array}---\left(4\right),$

In above formula, represent second mode { e _i, e _jall examples in corresponding normalization example antithesis table are to set, avg represents and asks calculation mean value;

Sub-step 4: by formula (5), ask calculation pattern c={e _i... e _kthe popularity degree calculated based on distance feedback function:

Reward _c＝min _i＜j≤k{Reward _c(e _i,e _j)}(5)，

Being judged to be final by being greater than to the candidate pattern of set a distance feedback threshold value θ | the same bit pattern in c| rank, namely final same bit pattern need meet Reward _c> θ;

Sub-step 5: with | c| rank with bit pattern for foundation, by front | the same bit pattern that c|-1 position has same characteristic features is asked also, obtains | c|+1 rank candidate peers pattern c ', makes c=c '.

Method and apparatus embodiment provided by the invention, compared with coordination mode excavation method of the prior art, advantage is enjoyed: first in Regional suitability, high efficiency and validity, do not need user's preset distance threshold value, but from the range distribution rule of the example in region example and its coverage, show that corresponding distance Truncation Parameters carrys out alternative distances threshold value, therefore excavate and be not limited to the degree of understanding of user to data, there is stronger region applicability; The second, owing to employing Thiessen polygon, make to connect calculating and be limited in the coverage of example, instead of the connection thinking that traditional single distance threshold limits, therefore significantly reduce connection calculated amount, improve the efficiency of excavation; 3rd, consider the distance weighting of instant teaching, respected First Law of Geography, be therefore a kind of coordination mode excavation method more geared to actual circumstances, cogency is also stronger.Above-described content can combine enforcement individually or in every way, and these variant are all within protection scope of the present invention.

Last it is noted that above embodiment is only in order to illustrate technical scheme of the present invention, be not intended to limit.Although with reference to previous embodiment to invention has been detailed description, those of ordinary skill in the art is to be understood that: it still can be modified to the technical scheme described in foregoing embodiments, or carries out equivalent replacement to wherein portion of techniques feature; And these amendments or replacement, do not make the essence of appropriate technical solution depart from the spirit and scope of various embodiments of the present invention technical scheme.

Claims (10)

1. take a self-adaptation coordination pattern acquiring method for space instances distance weighting into account, it is characterized in that, comprising:

Using the example of the same space feature as target layer, build Thiessen polygon, based on example and the further feature example of current goal layer in the annexation of Thiessen polygon coverage, structure normalization example antithesis table;

Distribution parameter calculating is carried out to the distance set of normalization example antithesis table, obtains normalized cumulant Truncation Parameters;

Utilize normalized cumulant Truncation Parameters to construct distance feedback function, calculate the distance backoff values that the connection example of any two different characteristics is right;

Under described distance feedback threshold value, obtain the final same bit pattern calculated based on distance feedback function.

2. method according to claim 1, is characterized in that, described based on Thiessen polygon, and the step building normalization example antithesis table comprises:

For each space-like feature e _i, using the example as target layer, create Thiessen polygon figure;

Between the zone of influence of the object instance of delimiting based on Thiessen polygon figure, set up each example with further feature e _jthe incidence relation of example, and these examples associated are stored in two-dimentional Hash table table_instance to the distance with them _vorin; Wherein, table_instance _vor(e _i, e _j) be the orderly pattern (e of second order _i, e _j) Cell information, represent with e _ifor target layer, be characterized as e _jexample carry out being connected obtained row example information with target layer example, representation feature is e _ia kth example, Cell information comprises the orderly pattern (e of second order _i, e _j) connected example to the distance value dis of insPair and its correspondence;

By formula (1), calculate different characteristic connection example between normalized cumulant, replace the distance value of described two-dimentional Hash table:

$Nordis(p_k^{e_i},p_w^{e_j})=\frac{dis(p_k^{e_i},p_w^{e_j})-\min \left({\mathrm{Dis}}_{VP}\right)}{\max \left({\mathrm{Dis}}_{VP}\right)-\min \left({\mathrm{Dis}}_{VP}\right)}---\left(1\right),$

Wherein, dis (a, b) represents two example a, the Euclidean distance between b,

Dis _vPrepresent table_instance _vorin the right distance set of all example,

Min (T) and max (T) represents the minimum and maximal value of getting set T respectively;

Unordered process is carried out to described two-dimentional Hash table, obtains normalization example antithesis table step comprises:

The table_instance meeting j>i _vor(e _j, e _i) in the example of arbitrarily in order second mode Cell information order is exchanged after, obtain new two-dimentional Hash table

Meet j>i's by described two-dimentional Hash table newly cell information and former table_instance _vor(e _i, e _j) corresponding Cell information carries out asking and processing, and obtains final normalization example antithesis table

3. method according to claim 1, is characterized in that:

Instance number is greater than to the data set of 12000, utilizes the normalized cumulant set in described normalization example antithesis table, under the prerequisite of level of signifiance α=0.05, carry out the parameter fitting of broad sense Extreme maximum distribution function, for χ ²the situation that in inspection, the goodness of fit is greater than 0.8, the normalized cumulant described in setting blocks as μ ± stderr; Wherein μ represents the location parameter of generalized extreme value maximum value fitting function, and stderr represents the residual error of matching; Or,

Other is not met to the situation of instance number and goodness of fit condition, using the average of normalized cumulant as cutoff value.

4. method according to claim 1, is characterized in that, utilizes described normalized cumulant Truncation Parameters, builds distance feedback function, calculates the connection example pair of any two different characteristics distance backoff values:

$\mathrm{Re}ward(p_k^{e_i},p_w^{e_j})=\left\{\begin{array}{cc}{\left(\frac{l_{close}-Nordis(p_k^{e_i},p_w^{e_j})}{l_{close}}\right)}^2& Nordis(p_k^{e_i},p_w^{e_j})-l_{close}\\ -{\left(\frac{Nordis(p_k^{e_i},p_w^{e_j})-l_{far}}{1-l_{far}}\right)}^2& Nordis(p_k^{e_i},p_w^{e_j})>l_{far}\\ 0& l_{close}\&le;Nordis(p_k^{e_i},p_w^{e_j})\&le;l_{far}\end{array}\right.---\left(2\right),$

Wherein, l _close=μ-stderr, and l _far=μ+stderr.

5. method according to claim 1, is characterized in that, under described distance feedback threshold value, the step obtaining the final same bit pattern based on the calculating of distance feedback function comprises:

With the feature corresponding to the column locations of described normalization example antithesis table to { e _i, e _jbe initial candidate peers pattern c, the following sub-step 1-5 of iteration, until | c| rank are not greater than the member of distance feedback threshold value with bit pattern, wherein, and j>i,

When | during c|=2, omit sub-step 1-2, directly jump to sub-step 3;

Sub-step 1: for candidate pattern c={e _i... e _k, according to the example pair in described normalization example antithesis table, obtain the group example insClique of candidate pattern _c, wherein, the feature comprised in candidate pattern c arranges according to lexcographical order, namely meets k>=j>i, insClique _crefer to based on Thiessen polygon, spatially have the combination of the example of the different characteristic of full annexation, namely roll into a ball any two examples pair extracted in example, can respective items be found in normalization example antithesis table;

Sub-step 2: according to formula (3), asks any two the second mode { e calculating and comprise in candidate pattern c _i, e _jexample pair at insClique _cdistance backoff values under constraint:

${\mathrm{Reward}}_c(p_k^{e_i},p_w^{e_j})=\left\{\begin{array}{cc}\mathrm{Re}ward(p_k^{e_i},p_w^{e_j})& (p_k^{e_i},p_w^{e_j})\&Element;{\mathrm{\&pi;}}_{\{e_i,e_j\}}\left({\mathrm{insClique}}_c\right)\\ -\left|\right|\mathrm{Re}ward(p_k^{e_i},p_w^{e_j})\left|\right|& (p_k^{e_i},p_w^{e_j})\&NotElement;{\mathrm{\&pi;}}_{\{e_i,e_j\}}\left({\mathrm{insClique}}_c\right)\end{array}\right.---\left(3\right),$

Wherein, represent and ask for second mode { e _i, e _jat insClique _con projection;

Sub-step 3: by formula (4), asks and calculates second mode { e _i, e _jat c={e _i... e _kconstraint under distance backoff values:

$\begin{array}{c}{\mathrm{Reward}}_c(e_i,e_j)=\\ avg\left\{{\mathrm{Reward}}_c(p_k^{e_i},p_w^{e_j})\right|(p_k^{e_i},p_w^{e_j})\&Element;table\_{\mathrm{instance}}_{\stackrel{\&OverBar;}{vor}}(e_i,e_j).insPair\}\end{array}---\left(4\right),$

Wherein, represent second mode { e _i, e _jall examples in corresponding normalization example antithesis table are to set, avg represents and asks calculation mean value;

Sub-step 4: by formula (5), ask calculation pattern c={e _i... e _kthe popularity degree calculated based on distance feedback function:

Reward _c＝min _i＜j≤k{Reward _c(e _i,e _j)}(5)，

Being judged to be final by being greater than to the candidate pattern of set a distance feedback threshold value θ | the same bit pattern in c| rank, namely final same bit pattern need meet Reward _c> θ;

Sub-step 5: with | c| rank with bit pattern for foundation, by front | the same bit pattern that c|-1 position has same characteristic features is asked also, obtains | c|+1 rank candidate peers pattern c ', makes c=c '.

6. take a self-adaptation coordination pattern acquiring device for space instances distance weighting into account, it is characterized in that, comprising:

Normalization example antithesis table builds module, for using the example of the same space feature as target layer, build Thiessen polygon, based on example and the further feature example of current goal layer in the annexation of Thiessen polygon coverage, structure normalization example antithesis table;

Normalized cumulant Truncation Parameters acquisition module, for carrying out distribution parameter calculating to the distance set of normalization example antithesis table, obtains normalized cumulant Truncation Parameters;

Example is adjusted the distance backoff values computing module, for utilizing normalized cumulant Truncation Parameters to construct distance feedback function, calculates the distance backoff values that the connection example of any two different characteristics is right;

Coordination pattern acquiring module, under described distance feedback threshold value, obtains the final same bit pattern calculated based on distance feedback function.

7. device according to claim 6, is characterized in that, described normalization example antithesis table build module specifically for:

Thiessen polygon creating unit, for for each space-like feature e _i, using the example as target layer, create Thiessen polygon figure;

Two dimension Hash table creating unit, between the zone of influence of object instance delimited based on Thiessen polygon figure, sets up each example with further feature e _jthe incidence relation of example, and these examples associated are stored in two-dimentional Hash table table_instance to the distance with them _vorin; Wherein, table_instance _vor(e _i, e _j) be the orderly pattern (e of second order _i, e _j) Cell information, represent with e _ifor target layer, be characterized as e _jexample carry out being connected obtained row example information with target layer example, representation feature is e _ia kth example, Cell information comprises the orderly pattern (e of second order _i, e _j) connected example to the distance value dis of insPair and its correspondence;

Normalized cumulant computing unit, for the connection example that calculates different characteristic between normalized cumulant, replace the distance value of described two-dimentional Hash table, see formula (1):

$Nordis(p_k^{e_i},p_w^{e_j})=\frac{dis(p_k^{e_i},p_w^{e_j})-\min \left({\mathrm{Dis}}_{VP}\right)}{\max \left({\mathrm{Dis}}_{VP}\right)-\min \left({\mathrm{Dis}}_{VP}\right)}---\left(1\right),$

Wherein, dis (a, b) represents two example a, the Euclidean distance between b,

Dis _vPrepresent table_instance _vorin the right distance set of all example,

Min (T) and max (T) represents the minimum and maximal value of getting set T respectively;

Unordered processing unit, for carrying out unordered process to described two-dimentional Hash table, obtains normalization example antithesis table

Described unordered processing unit, specifically for:

The table_instance meeting j>i _vor(e _j, e _i) in the example of arbitrarily in order second mode Cell information order is exchanged after, obtain new two-dimentional Hash table

Meet j>i's by described two-dimentional Hash table newly cell information and former table_instance _vor(e _i, e _j) corresponding Cell information carries out asking and processing, and obtains final normalization example antithesis table

8. device according to claim 6, is characterized in that, described normalized cumulant Truncation Parameters acquisition module specifically for:

Instance number is greater than to the data set of 12000, utilizes the normalized cumulant set in described normalization example antithesis table, under the prerequisite of level of signifiance α=0.05, carry out the parameter fitting of broad sense Extreme maximum distribution function, for χ ²the situation that in inspection, the goodness of fit is greater than 0.8, the normalized cumulant described in setting blocks as μ ± stderr; Wherein μ represents the location parameter of generalized extreme value maximum value fitting function, and stderr represents the residual error of matching; Or,

Other is not met to the situation of instance number and goodness of fit condition, using the average of normalized cumulant as cutoff value.

9. device according to claim 6, is characterized in that, described example adjust the distance backoff values computing module specifically for, calculate the connection example pair of any two different characteristics distance backoff values, see formula (2):

$\mathrm{Re}ward(p_k^{e_i},p_w^{e_j})=\left\{\begin{array}{cc}{\left(\frac{l_{close}-Nordis(p_k^{e_i},p_w^{e_j})}{l_{close}}\right)}^2& Nordis(p_k^{e_i},p_w^{e_j})-l_{close}\\ -{\left(\frac{Nordis(p_k^{e_i},p_w^{e_j})-l_{far}}{1-l_{far}}\right)}^2& Nordis(p_k^{e_i},p_w^{e_j})>l_{far}\\ 0& l_{close}\&le;Nordis(p_k^{e_i},p_w^{e_j})\&le;l_{far}\end{array}\right.---\left(2\right),$

Wherein, l _close=μ-stderr, and l _far=μ+stderr.

10. device according to claim 6, is characterized in that, described coordination pattern acquiring module, specifically for:

With the feature corresponding to the column locations of described normalization example antithesis table to { e _i, e _jbe initial candidate peers pattern c, the following sub-step 1-5 of iteration, until | c| rank are not greater than the member of distance feedback threshold value with bit pattern, wherein, and j>i,

When | during c|=2, omit sub-step 1-2, directly jump to sub-step 3;

Sub-step 1: for candidate pattern c={e _i... e _k, according to the example pair in described normalization example antithesis table, obtain the group example insClique of candidate pattern _c, wherein, the feature comprised in candidate pattern c arranges according to lexcographical order, i.e. k>=j>i, insClique _crefer to based on Thiessen polygon, spatially have the combination of the example of the different characteristic of full annexation, namely roll into a ball any two examples pair extracted in example, can respective items be found in normalization example antithesis table;

Sub-step 2: according to formula (3), asks any two the second mode { e calculating and comprise in candidate pattern c _i, e _jexample pair at insClique _cdistance backoff values under constraint:

${\mathrm{Reward}}_c(p_k^{e_i},p_w^{e_j})=\left\{\begin{array}{cc}\mathrm{Re}ward(p_k^{e_i},p_w^{e_j})& (p_k^{e_i},p_w^{e_j})\&Element;{\mathrm{\&pi;}}_{\{e_i,e_j\}}\left({\mathrm{insClique}}_c\right)\\ -\left|\right|\mathrm{Re}ward(p_k^{e_i},p_w^{e_j})\left|\right|& (p_k^{e_i},p_w^{e_j})\&NotElement;{\mathrm{\&pi;}}_{\{e_i,e_j\}}\left({\mathrm{insClique}}_c\right)\end{array}\right.---\left(3\right),$

Wherein, represent and ask for second mode { e _i, e _jat insClique _con projection;

Sub-step 3: by formula (4), asks and calculates second mode { e _i, e _jat c={e _i... e _kconstraint under distance backoff values:

$\begin{array}{c}{\mathrm{Reward}}_c(e_i,e_j)=\\ avg\left\{{\mathrm{Reward}}_c(p_k^{e_i},p_w^{e_j})\right|(p_k^{e_i},p_w^{e_j})\&Element;table\_{\mathrm{instance}}_{\stackrel{\&OverBar;}{vor}}(e_i,e_j).insPair\}\end{array}---\left(4\right),$

Wherein, represent second mode { e _i, e _jall examples in corresponding normalization example antithesis table are to set, avg represents and asks calculation mean value;

Sub-step 4: by formula (5), ask calculation pattern c={e _i... e _kthe popularity degree calculated based on distance feedback function:

Reward _c＝min _i＜j≤k{Reward _c(e _i,e _j)}(5)，

Being judged to be final by being greater than to the candidate pattern of set a distance feedback threshold value θ | the same bit pattern in c| rank, namely final same bit pattern need meet Reward _c> θ;

Sub-step 5: with | c| rank with bit pattern for foundation, by front | the same bit pattern that c|-1 position has same characteristic features is asked also, obtains | c|+1 rank candidate peers pattern c ', makes c=c '.